IT Glue Documentation Auditor

A worked example of an advanced workflow: a Claude-managed scheduled agent that, once a month, sweeps your IT Glue organizations, scores each one on two independent axes — documentation completeness and record freshness — ranks clients by their documentation debt, and publishes a ranked report to Slack as a canvas with a one-line summary. There are no servers and no code to deploy: the agent is a saved prompt plus a monthly schedule plus two MCP connectors, running in Claude's cloud.

What it builds

The finished workflow runs unattended on the 1st of each month. Each run it enumerates its scoped set of organizations, pulls documents, configurations, and flexible assets for each, and produces two scores: a completeness score (are the records there?) and a freshness score (have they been touched in the last year?). Those two axes combine into a per-org debt score. The run ranks orgs by debt and delivers a canvas report that turns silent documentation rot into a prioritized backlog your team can act on before it bites during an outage.

Prerequisites

Before building, your Claude account needs all of the following:

Known gotchas

These are the things that cost real time the first time through. Account for them up front and the build genuinely takes minutes.

• A large IT Glue tenant has many organizations — scope the routine to a fixed set for large tenants. Like the Datto fleet sweep, enumerating every org in a large tenant and then making three tool calls per org (documents, configurations, flexible assets) adds up fast. For tenants with more than ~30 organizations, bake a fixed set of ~15–20 representative client IDs into the routine prompt. For smaller tenants, the routine can call search_organizations dynamically each run. Widening scope is a routine prompt change — no code to deploy.
• Use list and search payloads — do not get records one at a time. search_documents, search_configurations, and search_flexible_assets all accept an organization_id and return a full list in one call. Calling a per-record get for every document or configuration item across every org risks the 60-second tool timeout and multiplies API calls needlessly. One search call per org per data type is all this routine needs.
• Grade coverage against expected asset types, not an absolute count. A small client legitimately has fewer flexible asset types than an enterprise. The routine calls list_flexible_asset_types once to get the full type catalogue, then judges each org against the types relevant to it — not against the total type count. Without this, every SMB client will look under-documented compared to a large org, and the debt ranking becomes meaningless.
• Freshness is a proxy for currency, not a guarantee of accuracy. IT Glue has no "is this information correct?" field. The routine uses updated_at timestamps on documents and flexible assets as a staleness signal: records not touched in 12 months are flagged. A record updated yesterday might still be wrong; a record updated 18 months ago might still be accurate. The score surfaces forgotten records, not errors — treat it as a prompt to review, not a verdict.
• permitted_tools must be populated per connector. A routine with a connector attached but an empty permitted_tools list runs with no tools and silently does nothing — no error, no output. List the exact tool names the routine needs for each connector.
• A routine reaches only its attached connectors. The routine sandbox blocks arbitrary network egress, so notifications must go through the Slack connector's tools, not an outbound webhook. Attach every connector the workflow touches.
• Faster-than-hourly cron is rejected, and cron is always UTC. This routine runs monthly — 0 12 1 * * fires at 12:00 UTC, which is 08:00 America/New_York. If you adjust the schedule, convert your local time to UTC before setting it.
• If Slack is not in the /schedule connector list, read its details from an existing Slack routine. Open an existing routine that already uses Slack (via RemoteTrigger list → get → mcp_connections) and copy the connector_uuid and url from there.

The one-shot build prompt

With the connectors above in place, paste this to Claude. It confirms the gateway, assesses your IT Glue tenant's scale, creates the routine, and verifies it end to end.

Build me a scheduled IT Glue Documentation Auditor agent. Do all of this end to end:

1. Confirm the WYRE MCP Gateway works and IT Glue is reachable: call
   search_organizations with no filters and verify it returns a list of
   organizations. Note the total count.
2. Assess scale. If the tenant has more than ~30 organizations, you will need to
   scope the routine to a fixed set of key orgs rather than sweeping every one.
   Large tenants produce too much per-run payload to complete inside the tool
   timeout. Pick a representative set of about 15-20 organizations that covers
   the full client base (enterprise, mid-market, SMB) and record their IT Glue
   organization IDs. These IDs get baked into the routine prompt. For smaller
   tenants, the routine can enumerate all orgs dynamically.
3. Call list_flexible_asset_types to see the full set of asset types configured
   in this IT Glue tenant. Note the type names and IDs — the routine uses these
   to grade asset coverage per org rather than expecting an absolute count.
4. Confirm a Slack connector is connected. Note the destination channel name
   and ID (e.g. #documentation or #sw-dev). If Slack does not show in the
   /schedule connector list, read its connector_uuid and url from an existing
   routine that already uses Slack (RemoteTrigger list -> get -> mcp_connections).
5. Create a Claude-managed scheduled routine named "IT Glue Documentation Auditor":
   - Schedule: monthly, cron "0 12 1 * *" (1st of each month 08:00
     America/New_York = 12:00 UTC). Faster-than-hourly cadences are rejected.
     Cron is always expressed in UTC.
   - Attach TWO connectors, each with permitted_tools populated:
       * WYRE MCP Gateway: search_organizations, search_documents,
         search_configurations, search_flexible_assets, list_flexible_asset_types
       * Slack: slack_create_canvas, slack_send_message
     An empty permitted_tools list = the routine runs with no tools.
   - Routine prompt: every run, enumerate the scoped organizations, pull
     documents/configurations/flexible-assets for each, score completeness and
     freshness on two independent axes, rank orgs by documentation debt, publish
     a canvas titled "Documentation Audit — <month YYYY-MM>", and post a one-line
     summary to the channel. Use the exact routine prompt below, substituting the
     baked-in org IDs and channel ID.
6. Trigger a manual run and verify: a canvas titled "Documentation Audit —
   <current month>" was created with a per-org ranked debt list, a one-line
   summary landed in the destination channel, and the org count in the report
   matches the scoped set (spot-check one org's document count against IT Glue).

The resulting routine prompt

This is the lean prompt the build process installs into the scheduled routine itself. Substitute your own scoped org IDs (or remove the fixed list for small tenants), your staleness threshold if different from 12 months, and your destination channel.

You are the IT Glue Documentation Auditor. You run monthly on the 1st. Keep it lean.

Use ONLY: search_organizations, search_documents, search_configurations,
search_flexible_assets, list_flexible_asset_types,
slack_create_canvas, slack_send_message.

ORGANIZATIONS TO AUDIT:
<If scoped to a fixed set, list org IDs and names here, e.g.:>
- <org-id-1> - <Client Name 1>
- <org-id-2> - <Client Name 2>
- ... ~15-20 orgs ...
<If the tenant is small enough to enumerate dynamically, call
search_organizations with no filter and use the returned list.>

STALENESS THRESHOLD: Any document or password record not updated in the last
12 months is considered STALE.

PHASE 1 - Enumerate expected asset types.
Call list_flexible_asset_types once. Record all active flexible asset type
names and IDs. This is the reference set used in Phase 2 step 4.

PHASE 2 - Per-organization audit.
For each organization in the scoped list, run these steps. If a call fails,
record the org as "could not read" and continue.

  Step 1 — Documents (completeness + freshness).
  Call search_documents with the organization_id. For each document, read:
  name, updated_at. Count total documents. Flag any document where updated_at
  is more than 12 months ago as STALE.

  Step 2 — Configurations (coverage).
  Call search_configurations with the organization_id. Count total
  configuration items. A configuration item with no linked documents is an
  UNDOCUMENTED CONFIG. Read only: id, name, organization_id.

  Step 3 — Flexible assets (coverage by type).
  Call search_flexible_assets with the organization_id. Count how many of the
  expected asset types (from Phase 1) are represented in this org's flexible
  assets. A missing type is a COVERAGE GAP — but only flag it as a gap if the
  org is a managed client where that type is expected. Small / basic orgs
  legitimately have fewer asset types; do not penalize them for types that do
  not apply.

  Step 4 — Score the org.
  Assign two scores, each 0–100:
  - COMPLETENESS score: based on (a) whether documents exist at all,
    (b) ratio of documented configs to total configs, and (c) fraction of
    expected flexible asset types present.
  - FRESHNESS score: based on the percentage of documents and flexible assets
    updated within the last 12 months. An org with no stale records scores 100.
    An org with all stale records scores 0.
  Compute a DEBT score = 100 - ((completeness + freshness) / 2). Higher debt = worse.

PHASE 3 - Rank and build the report.
Sort organizations by DEBT score descending (highest debt first).
Compose a markdown report:
- SUMMARY at the top: total orgs audited, number with completeness < 50,
  number with freshness < 50, number of "could not read" orgs.
- RANKED ORG LIST: one section per org, ordered by debt, showing:
    org name | completeness score | freshness score | debt score
    - Documents: total, stale count
    - Configs: total, undocumented count
    - Flexible assets: types present / types expected, list any gaps
- COULD NOT READ: list any orgs that could not be read this run.

PHASE 4 - Deliver.
Call slack_create_canvas titled "Documentation Audit — <YYYY-MM for current month>"
with the full ranked report as markdown.
Then call slack_send_message to #<your-channel> (channel ID <your-channel-id>):
a one-line summary, e.g.:
"Documentation audit: <N> orgs, avg debt <X>. Top offender: <org name>
(debt <score>). See canvas: <link>"

If every org scored completeness >= 80 AND freshness >= 80, send only:
"Documentation audit: all <N> orgs in good shape this month (completeness >=80,
freshness >=80)." and skip the canvas.

How it works

A monthly cadence, by design

Documentation debt accumulates slowly — runbooks go stale over months, not minutes, and missing records stay missing until an engineer notices during an incident. A monthly audit is frequent enough to catch drift before it causes harm, and infrequent enough to keep the signal meaningful: a team seeing the same report every week tends to tune it out. Running on the 1st of each month gives the report a natural rhythm that ties into monthly reviews and client QBRs.

Completeness and freshness as two distinct axes

The routine scores each organization on two independent dimensions. Completeness asks whether the records exist: are there documents at all, what fraction of configuration items have linked documentation, and how many of the expected flexible asset types are represented? Freshness asks whether the records have been touched recently: what percentage of documents and flexible assets were updated within the last 12 months? The two scores combine into a debt score, but they are reported separately so the fix is obvious — an org with high completeness and low freshness needs a review pass, while an org with low completeness and high freshness needs new records created, not just updated.

Grading against expected asset types, not absolute counts

IT Glue's flexible asset catalogue can contain dozens of types — network diagrams, backup configs, security baselines, vendor contacts, and more. A small client with five employees legitimately has only a handful of types configured; grading them against the full enterprise type catalogue would make every SMB look critically under-documented. The routine calls list_flexible_asset_types once per run to retrieve the full type catalogue, then grades each org against the subset of types relevant to its size and service tier. That relative grading is what keeps the ranked debt list actionable: an org near the top has genuinely missing documentation for its own context, not just fewer types than a larger client.

Canvas and summary delivery

The per-org ranked report can be long — one section per client with three data points each — so the routine publishes it as a Slack canvas, a durable document teammates can scroll and reference month to month, and posts only a one-line summary to the channel with the top offender and a link to the canvas. When every org scores well (completeness and freshness both ≥ 80), it posts a single clean line and skips the canvas entirely. See delivery adapters for other ways to surface a routine's output.

Extending it

The natural pairing is Hudu: if you run Hudu alongside IT Glue, a runbook-freshness-auditor routine built on Hudu's API surfaces runbook-specific currency — the same freshness signal applied specifically to runbooks and SOPs rather than all documentation. The two routines make a complementary pair: one covers the full IT Glue documentation estate, the other focuses on the procedural layer that is most critical during an incident.

A second extension connects the audit to the onboarding flow: documentation gaps found during client onboarding — asset types not yet populated, configs not yet linked — can be fed into an onboarding-completeness-checker portfolio agent as a baseline, so the monthly audit measures progress against a known starting state rather than an arbitrary threshold.

Questions or a workflow you'd like documented? Open an issue in the msp-claude-plugins repository.